When using Kagi Assistant with model set to Qwen3.6 Plus, the output prompt is put in the thinking block.
[upl-image-preview uuid=01ac7644-c9ef-4a9e-bb0a-b749a7114a42 url=
alt={TEXT?}
This does not appear to happen with other models.
Expecting response formatted similar to the Kimi K2.5 example.
